Statistical analysis and modelling of bi-modal autofluorescence-Raman imaging for efficient diagnosis and treatment of biological tissues
Lead Research Organisation:
Royal Holloway University of London
Department Name: Mathematics
Abstract
Raman spectroscopy, a particular type of spectroscopic technique to measure molecular vibrations, has been successfully applied to determine chemical composition of biological tissues at sufficiently fine scales. This has in turn allowed the acquired information to be readily used for diagnosis and surgical treatment of skin, and potentially other types of cancer, such as breast cancer. The approach promises a more accurate and significantly less costly alternative to the existing diagnosis and surgical practices, and therefore a more rapid and broader access to these types of healthcare. In terms of patient experience the approach also offers an improvement through a significant reduction of both, the amount of unnecessarily removed or otherwise traumatised tissue, and the lag between successive stages of the procedure. In its naïve implementation, the method would first scan the entire biological sample before processing the spectral data from each site for subsequent automated analysis to establish presence or absence of cancerous formations in the sample or, more ambitiously, to produce a biological description at each probed location. However, this naïve implementation takes prohibitively long time, defeating the purpose of applying the method during a single uninterrupted surgery. A bi-modal imaging solution has been proposed, in which a nearly instantaneous preliminary autofluorescence imaging combined with an automated clustering technique subsequently guides the Raman spectrometer to concentrate its measurements on segments deemed more likely to contain cancer. A statistical model trained on a large number of previously analysed samples then attempts to complete the cancer detection task in each segment, requesting more Raman measurements if it is not already sufficiently confident in the current diagnosis. A device implementing this methodology has now been trialed in one, and is ready to be trialed in other NHS centres, as well as internationally. While the currently reported results produced by the current implementation of the methodology are encouraging, there still remain several directions for advancing the methodology and subsequently transforming the existing technology to a cutting edge final product that will meet the expectations of the healthcare providers and patients. In particular, while the objective evaluation in the hospital largely confirms the expected sensitivity of the method as 90%, the specificity (proportion of non-cancer samples correctly diagnosed as non-cancer) is notably lower than expected. Also, while currently the technology requires upto 30 minutes to produce a diagnosis, the ultimate aim is 5-10 minutes. We propose advanced statistical and computational models and methods, which utilize previously under-utilized spatial and morphological information to achieve the required transformation. The proposed methods also include upgrading currently used generic Multivariate Statistical Analysis to Functional Data Analysis, which takes advantage of the intrinsic functional structure of the Raman spectra and hence extracts more accurate biochemical markers from the analysed biological tissues. Methods of non-Euclidean statistics are also considered to efficiently capture and represent spatial variation in the spectral data and subsequently use such spatial information for more accurate recognition of the tissue types. The recent enlargement of the available data shall also allow us to take advantage of more complex statistical classification models capturing finer differences between the tissue types and subsequently leading to more accurate and robust detection of cancer. The proposal is also supported by our research partners from Estonia (EU) who are going to complement our work by investigating an additional class of statistical models, increasing the overall chance of delivering a highly valuable final product.
| Description | The award has now finished but the research originating from the award is ongoing, hence the findings reported below are likely to continue to evolve. In the course of transforming our prototype method that automates diagnosis and treatment of skin cancer into a real healthcare product, we have discovered that the first stage of the medical imaging underpinning our method may contain currently underutilized information relevant for meeting our clinical objectives. The existing method has been designed to detect Basal Cell Carcinoma (BCC), the most common type of cancer in humans, and relies on Autofluorescence (AF) imaging (first stage) to guide Raman microscopy (second stage) to obtain biochemical information from a human tissue for determining its type. The guidance is provided via an image segmentation technique aiming to minimise the risk of missing cancer. Using our earlier data, we found evidence that shape of an AF segment may be useful for 'trimming' 6-8% of segments that do not contain BCC, without compromising BCC detection. By thereby allowing the method to direct its Raman analysis, which is significantly more time consuming and costly than AF, toward more relevant regions in the tissue, the prospective elimination of irrelevant segments has the potential to improve the overall accuracy of the method without extending the analysis time. Our findings are obtained using a prominent method of Topological Data Analsyis, called Smooth Weighted Euler Characteristic Transform (SWECT). SWECT was recently proposed (by others) as a transformation to represent shape of an image segment by a finite set of numerical matrices. Each matrix in the set represents the segment under a distinct rotation, and equals any other matrix in the set when its columns are circularly permuted. To compare shapes and to relate them to tissue types, we have used an appropriate rotation invariant counterpart of the Euclidean distance on matrices in combination with advanced Statistical and Machine Learning techniques. There, we have learned of an important limitation of the highly successful kernel methods, such as Support Vector Machines, when dealing with non-Euclidean distances, such as the above rotation invariant distance. We have identified practical ways to circumvent this limitation, which may also be of relevance to statistical data analysis on other non-Euclidean spaces. Also, following the general approach (of Fréchet and Karcher) to non-Euclidean statistics, we have extended the notion of sample mean and sample variance to the resulting metric space of shapes. Subsequently, we have adapted the well-known Generalized Procrustes Analysis algorithm to compute SWECT of mean shape of a sample of shapes, as well as the sample variance. This has in turn made more classification and clustering algorithms, e.g. k-means and agglomerative clustering, available to us for clustering and also for supervised classification of AF segments by their shape. These results have also allowed us to better visualize shape segments using prominent dimensionality reduction approaches. By analyzing results from applying these algorithms to our data sets collected at different stages of our project, we have found evidence of certain discrepancies between the datasets. A more thorough data analysis has suggested that two changes are likely to have taken place in the data acquisition or processing (AF segmentation) that need to be fully understood before making necessary adjustments to the classification algorithms. This is the subject of our ongoing work. While the current implementation of the AF-Raman method bases its final classification of the tissue solely on the biochemical characteristics captured by the Raman spectroscopy, additional inclusion of morphological characteristics is hypothesised to hold the potential to improve the classification accuracy. Thus, we also developed tools for investigating if shape of AF segments could serve as as a morphological signature of tissue types. Specifically, these tools allow us to use shape in conjunction with the Raman spectroscopic features within a range of statistical learning algorithms that can be trained to classify new tissue types. These tools also allow us to investigate if the combined use of shape of AF segments and Raman features results in improved performance of the algorithm. The core of these tools is a `Support Measure Machine' (SMM) developed in close cooperation with our Estonian partner. In addition to the above statistical shape analysis, SMM allows us to incorporate spatial dependence into our AF-Raman technology, which was a key objective of this award. |
| Exploitation Route | Our research outcomes are an integral part of the larger ongoing multidisciplinary collaboration led by Professor Ioan Notingher (University of Nottingham, Co-Investigator on this award), involving medical and industrial biotechnology partners, aiming to deliver an efficient healthcare product. These findings are immediately relevant in numerous applications where image segmentation is a key component, as we have demonstrated viability of the methodology that combines Euler Curve Transforms (developed by others) with statistical shape analysis and statistical and machine learning for using shape as predictor of a certain condition or disease. Specifically, the currently reported `coordinate-free' approach to non-Euclidean statistical analysis and machine learning of morphological (shape) data promotes the recently proposed (by others) Weighted Euler Characteristic Transform as a particularly useful method of Topological Data Analysis, offering an alternative to landmark and diffeomorphism-based approaches, at the same time raising questions of theoretical and practical nature (e.g. computation of the inverse transform). These questions are of interest to applied statisticians, computer scientists, and mathematicians working in Topological and Geometric Data Analysis. While our own ongoing efforts are to extend our finding to the application of our Autofluoresence (AF)-Raman technology to the diagnosis and treatment of breast cancer, researchers using similar multimodal approaches may find our findings useful in their particular imaging domains. Indeed, the advancement of our bi-modal application of AF-Raman technology has led other biophysics teams to look into the multi-modal imaging approach. As our new results contribute to this advancement, the case for this approach becomes stronger, which is likely to inspire its wider applicability. Based on our adaptation of the Generalized Procrustes Algorithm, we have also described a greedy algorithm to implement the Ward (within cluster variance minimization) criterion with the non-Euclidean distance between shapes for shape clustering. While we have thus far used this with the agglomerative clustering in the context of transductive learning, the algorithm can readily be extended for wider use. Also, the greedy version could be replaced by one that attempts to find the global minimum, which is likely to be more challenging. |
| Sectors | Education Healthcare Manufacturing including Industrial Biotechology |
| URL | https://link.springer.com/chapter/10.1007/978-3-031-65723-8_33 |
| Description | By improving the reliability and efficiency of the bi-modal autofluorescence-Raman technology to diagnose skin cancer and thus significantly facilitate micrographic ('Mohs') skin surgeries, the findings from this award are contributing to the ongoing transformation of this technology from a prototype device, which is already being trialed at an NHS hospital, to a product that fully meets the expectations of the healthcare providers and their patients. Thus, our findings are making it more likely for the use of this technology to become a widely acceptable medical practice, making Mohs surgeries more widely and readily available. These findings are also facilitating our concurrent efforts to extend this technology to the diagnosis and treatment of breast cancer, as evident from our current National Institute for Health Research (NIHR) `Invention for Innovation call 28 B' Stage 2 grant application NIHR208400 `Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'. |
| First Year Of Impact | 2025 |
| Sector | Healthcare |
| Impact Types | Societal Economic |
| Title | Balls4DistanceBasedClassification |
| Description | This is both a data analysis technique and a classification algorithm to use with distance based data. To the best of our knowledge the closest but more complex (and therefore slower) method is the `norm ball classifier for one-class classification' https://doi.org/10.1007/s10479-021-03964-x. While our adaptation of the Generalized Procrustes Algorithm (see previous entry) does allow us to extend the balance condition from the `norm ball classifier for one-class classification', we do not attempt to reduce the number of balls, and use order ranking with leave-one-out cross validation instead of the optimization used in https://doi.org/10.1007/s10479-021-03964-x. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | This simple method/algorithm has provided a useful model-free benchmark for our experiments on statistical shape analysis for prediction of Basal Cell Carcinoma (BCC) from autofluorescence data. More importantly, it has helped us identify discrepancies between different batches of data, which is crucial for our decision on integration of AF-based data analyses in the bi-modal AF-Raman technology. |
| Title | GPA4SWECTs |
| Description | PI developed an adaptation of the Generalized Procrustes Analysis algorithm to compute SWECT representation of average shape and variance of a sample of shapes represented by SWECTs. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2023 |
| Provided To Others? | No |
| Impact | The algorithm is immediately applicable for clustering, e.g. k-means or agglomerative clustering, as well as for visualization of shape data via their ECT, WECT, or SWECT representations. It was already used in the group's recent and forthcoming publications, in particular it contributed to detection of disparities between different datasets (see, e.g. previous and subsequent entries). The development and use of the algorithm was already acknowledged in the recent publication (see doi below), and its description is to be included in a forthcoming publication. Its application with the breast cancer data is also planned in the team's most recent NIHR Invention for Innovation Call 28 B grant application NIHR208400 ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'', see previous entry. |
| Title | SWECTsBatch2and3gAndDistances |
| Description | Using unnormalized Smooth Weighted Euler Curve Transform (SWECT) developed by others, PI created datasets SWECTs 2 and SWECTs 3 storing transformed auto-fluorecsence (AF) segments produced by the team's AF-Raman technology. Specifically, the AF segments resulted from applying the AF-Raman technology to tissue specimens of patients undergoing Mohs surgery at the Nottingham NUH NHS Treatment Centre, Batch 2 (59 patients) giving rise to SWECTs 2 and Batch 3g (15 patients) - to SWECTs 3. Ethical approval was granted by the Health Research Authority (HRA) and Health and Care Research Wales (HCRW) (18/WM/0105). All participating patients have given written informed consent for participation in the study and the use of their deidentified, anonymised, aggregated data and their case details for publication. Patients were recruited randomly, regardless of sex, age or anatomical characteristics. PI also created a dataset/matrix of distances between all shapes represented in SWECT2 and SWECT3. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2023 |
| Provided To Others? | No |
| Impact | The produced datasets and models that use them (see subsequent entries) for BCC prediction are currently being considered for integration in the AF-Raman technology as part of the team's ongoing efforts to transform the existing AF-Raman prototype technology/device to one that would be broadly used by skin cancer surgeons in their daily practice. Furthermore, the developed statistical and machine learning models to use shape as predictor of tissue type are to be adapted and tested in the new context of applying the AF-Raman technology for breast cancer diagnosis and treatment. Thus the above data and models contributed to the team's most recent NIHR Invention for Innovation Call 28 B grant application NIHR208400 ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'', which has already undergone the second stage of review, awaiting a final decision. |
| Title | SWECTsBatchDTADistances |
| Description | Unnormalized Smooth Weighthed Euler Curve Transforms of autofluorescence segments produced by the AF-Raman technology on Batch DTA data (112 patients), SWECTs DTA. Tissue specimens used in this work were obtained from patients undergoing Mohs surgery at the Nottingham NUH NHS Treatment Centre. Ethical approval was granted by the Health Research Authority (HRA) and Health and Care Research Wales (HCRW) (18/WM/0105). All participating patients have given written informed consent for participation in the study and the use of their deidentified, anonymised, aggregated data and their case details for publication. Patients were recruited randomly, regardless of sex, age or anatomical characteristics. The rotation invariant distances between each SWECT from the union of SWECTs 2, SWECTs 3, and SWECTs DTA, are stored in a related dataset/matrix. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | No |
| Impact | The creation of these datasets allowed us to extend our statistical and machine learning models, which were previously developed and tested on Batch 2 and Batch 3g (SWECTs 2, SWECTs 3, respectively) data, to these most recently collected clinical data. As a result, certain discrepancies, which emerged when comparing Batch 2 with Batch 3g, have now been fully confirmed as well as another discrepancy between Batches 2 and 3g on the one hand and more recent Batch DTA on the other, has been observed. By carefully investigating the nature of the discrepancies and making any necessary changes to the shape classification algorithms, we would ensure that the AF image acquisition process and AF segmentation algorithms have from now on been fixed, and that the data acquired up to date have been used maximally. |
| Title | SupervisedAndSemisupervisedShapeClassifiers |
| Description | Using the distance data (reported in two of the other entries) and the tissue types, i.e. Basal Cell Carcinoma (BCC) /other (non-BCC), of the given AF segments/shapes, as well as several existing statistical and machine learning methods to classify ``coordinate-free'' data directly from their distances, PI developed models, which, when trained on subsets of the above shape/tissue type data, predict the tissue type of the remaining shape data. PI cross-validated the models measuring their specificity (percentage of correctly predicted non-BCCs) while controlling their sensitivity (percentage of correctly detected BCCs) at 100% (as required by the given application). The earlier report (doi given below) uses Batch 2 for training and Batch 3g as a test set, based on which PI chose the best performing model for use with new data (Batch DTA, report is subject of forthcoming publication and presentations). Further, PI also developed a transductive agglomerative clustering shape classifier, to appear in the forthcoming publication and presentations (not covered by the report with the doi below). |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | These models and algorithms have demonstrated viability of the methodology to combine Euler Curve Transform (particularly SWECT) with the appropriate non-Eucledian statistical shape analysis and machine learning methods allowing shape (of segments) to be used as a predictor of a distinct response, condition, e.g. disease. Besides strengthening the recent reports (of others) on the suitability of ECT topological data analysis, these models and algorithms are additionally readily applicable to a wide class of imaging tasks, e.g. multimodal imaging applications involving image segmentation. |
| Description | RHUL-Nottingham-Tartu |
| Organisation | University of Nottingham |
| Department | School of Physics and Astronomy |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | I have researched existing methods and developed and implemented (in Matlab) new methods to analyze shape of auto-fluorescence segments. Using these methods, I have carried out statistical analysis to assess shape as a potential predictor of basal cell carcinoma. I was the lead author of our recent publication reporting on these developments, and I also presented them at two international workshops, and have also been accepted to present a poster at another international workshop. I have also assessed distinct batches of relevant data for consistency, finding clear evidence of systemic/structural changes, which pose new questions with regard to optimal integration of all these data in the AF-Raman technology previously developed by our team (prior to its recent inclusion of the Estonian partners). I have also coordinated involvement of the Estonian partners in our collaboration, making two research visits to the University of Tartu, and co-supervising a Master's project on exploratory analysis and cleaning of Raman spectroscopic data. I have also implemented a prototype of a `Support Measure Machine', which was developed jointly by an Estonian partner and myself (as part of this collaboration) to perform statistical classification of tissue types on variable size sets of Raman measurements. I have also contributed by seeking further funding for our collaboration, and in particular I am a Co-Investigator on the NIHR grant application ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'' led by the Nottingham collaborator. |
| Collaborator Contribution | The Nottingham collaborators have continued to acquire and process new patient data, engaging with medical colleagues to assess performance of our technology for diagnosis and treatment of skin cancer, and also extending to breast cancer. They have also led efforts to report the results in leading journals (1) https://doi.org/10.1002/jvc2.336, (2) https://doi.org/10.1007/s10549-024-07349-z, (3) https://doi.org/10.1093/bjd/ljae196. They have also contributed by providing the data to the other collaborators, and also by assessing and interpreting results of the statistical shape analysis of auto-fluorescence segments. They have also led the joint grant application ``Integrated auto-fluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'' awaiting the final stage decision from NIHR. They have also been reporting on our collaboration at conferences, e.g. an invited talk at https://clirpath-ai.org/clirpath-ai-conference-2025/. The Estonian collaborators have played a key role in developing the Support Measure Machine approach to classifying variable size sets of Raman measurements, which addresses the challenge of incorporating spatial dependence in the AF-Raman technology. An Estonian partner was also the lead supervisor of a Master's dissertation on exploratory analysis and cleaning of Raman spectral data. |
| Impact | This collaboration is multi-disciplinary, involving Mathematics and Statistics, Machine Learning/AI, Physics, Medical Sciences/Healthcare. Up to date: 1. Koloydenko, A.A., Notingher, I., Boitor, R., Lember, J. (2024). Shape Analysis of AF Segments for Rapid Assessment of Mohs Layers for BCC Presence by AF-Raman Microscopy. In: Einbeck, J., Maeng, H., Ogundimu, E., Perrakis, K. (eds) Developments in Statistical Modelling. IWSM 2024. Contributions to Statistics. Springer, Cham. https://doi.org/10.1007/978-3-031-65723-8_33 2. Contributed talk at an international workshop on Statistical Modelling: https://maths.dur.ac.uk/iwsm2024/files/iwsmDurham_1.pdf 3. Invited talk at an international workshop on Markov Modelling: https://math.ut.ee/en/content/tartu-workshop-markov-modelling 4. Joint supervision of Master's thesis: https://dspace.ut.ee/items/8922ebc2-4f61-4aea-907e-e3b0fbe982c9 5. Funding application, PI: The British Skin Foundation's Small Grant Scheme, unsuccessful. 6. Funding aplication, CoI: National Institute for Health and Care Research ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'', first round passed, awaiting decision from the second/final round. Forthcoming: 1. Contributed poster at an international workshop on Uncertainty in multivariate, non-Euclidean, and functional spaces: theory and practice https://www.newton.ac.uk/event/rclw01/, 6-9th May, 2025 2. Contributed poster at CLIRPath-AI: A roadmap for AI-based spectral pathology (EPSRC Healthcare Technologies NetworkPlus), 20-22nd May, 2025 |
| Start Year | 2022 |
| Description | RHUL-Nottingham-Tartu |
| Organisation | University of Tartu |
| Department | Institute of Mathematics and Statistics |
| Country | Estonia |
| Sector | Academic/University |
| PI Contribution | I have researched existing methods and developed and implemented (in Matlab) new methods to analyze shape of auto-fluorescence segments. Using these methods, I have carried out statistical analysis to assess shape as a potential predictor of basal cell carcinoma. I was the lead author of our recent publication reporting on these developments, and I also presented them at two international workshops, and have also been accepted to present a poster at another international workshop. I have also assessed distinct batches of relevant data for consistency, finding clear evidence of systemic/structural changes, which pose new questions with regard to optimal integration of all these data in the AF-Raman technology previously developed by our team (prior to its recent inclusion of the Estonian partners). I have also coordinated involvement of the Estonian partners in our collaboration, making two research visits to the University of Tartu, and co-supervising a Master's project on exploratory analysis and cleaning of Raman spectroscopic data. I have also implemented a prototype of a `Support Measure Machine', which was developed jointly by an Estonian partner and myself (as part of this collaboration) to perform statistical classification of tissue types on variable size sets of Raman measurements. I have also contributed by seeking further funding for our collaboration, and in particular I am a Co-Investigator on the NIHR grant application ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'' led by the Nottingham collaborator. |
| Collaborator Contribution | The Nottingham collaborators have continued to acquire and process new patient data, engaging with medical colleagues to assess performance of our technology for diagnosis and treatment of skin cancer, and also extending to breast cancer. They have also led efforts to report the results in leading journals (1) https://doi.org/10.1002/jvc2.336, (2) https://doi.org/10.1007/s10549-024-07349-z, (3) https://doi.org/10.1093/bjd/ljae196. They have also contributed by providing the data to the other collaborators, and also by assessing and interpreting results of the statistical shape analysis of auto-fluorescence segments. They have also led the joint grant application ``Integrated auto-fluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'' awaiting the final stage decision from NIHR. They have also been reporting on our collaboration at conferences, e.g. an invited talk at https://clirpath-ai.org/clirpath-ai-conference-2025/. The Estonian collaborators have played a key role in developing the Support Measure Machine approach to classifying variable size sets of Raman measurements, which addresses the challenge of incorporating spatial dependence in the AF-Raman technology. An Estonian partner was also the lead supervisor of a Master's dissertation on exploratory analysis and cleaning of Raman spectral data. |
| Impact | This collaboration is multi-disciplinary, involving Mathematics and Statistics, Machine Learning/AI, Physics, Medical Sciences/Healthcare. Up to date: 1. Koloydenko, A.A., Notingher, I., Boitor, R., Lember, J. (2024). Shape Analysis of AF Segments for Rapid Assessment of Mohs Layers for BCC Presence by AF-Raman Microscopy. In: Einbeck, J., Maeng, H., Ogundimu, E., Perrakis, K. (eds) Developments in Statistical Modelling. IWSM 2024. Contributions to Statistics. Springer, Cham. https://doi.org/10.1007/978-3-031-65723-8_33 2. Contributed talk at an international workshop on Statistical Modelling: https://maths.dur.ac.uk/iwsm2024/files/iwsmDurham_1.pdf 3. Invited talk at an international workshop on Markov Modelling: https://math.ut.ee/en/content/tartu-workshop-markov-modelling 4. Joint supervision of Master's thesis: https://dspace.ut.ee/items/8922ebc2-4f61-4aea-907e-e3b0fbe982c9 5. Funding application, PI: The British Skin Foundation's Small Grant Scheme, unsuccessful. 6. Funding aplication, CoI: National Institute for Health and Care Research ``Integrated autofluorescence-Raman spectroscopy (AF-Raman) for intra-operative assessment of sentinel lymph node biopsies in breast cancer surgery'', first round passed, awaiting decision from the second/final round. Forthcoming: 1. Contributed poster at an international workshop on Uncertainty in multivariate, non-Euclidean, and functional spaces: theory and practice https://www.newton.ac.uk/event/rclw01/, 6-9th May, 2025 2. Contributed poster at CLIRPath-AI: A roadmap for AI-based spectral pathology (EPSRC Healthcare Technologies NetworkPlus), 20-22nd May, 2025 |
| Start Year | 2022 |